Predicting and Understanding Emergency Shutdown Durations Level of Pipeline Incidents Using Machine Learning Models and Explainable AI

Pipeline incidents pose significant concerns due to their potential environmental, economic, and safety risks, emphasizing the critical need to understand and manage this vital infrastructure. While existing studies predominantly focus on the causes of pipeline incidents and failures, few have investigated the consequences, such as shutdown duration, and most lack comprehensive models capable of accurately predicting and providing actionable insights into the risk factors. This study bridges this gap by employing machine learning (ML) techniques, including Random Forest and Light Gradient Boosting Machine (LightGBM), for classifying pipeline incidents’ emergency shutdown duration levels. These techniques are specifically designed to capture complex, nonlinear patterns and interdependencies within the data, addressing the limitations of traditional linear approaches. The proposed model has further enhanced with Explainable AI (XAI) techniques, such as Shapley Additive exPlanations (SHAP) values, to improve interpretability and provide insights into the factors influencing shutdown durations. Historical incident data, collected from the Pipeline and Hazardous Materials Safety Administration (PHMSA) from 2010 to 2022, were utilized to examine the risk factors. K-Fold Cross-Validation with 5 folds was employed to ensure the model’s robustness. The results demonstrate that the LightGBM model achieved the highest accuracy of 75.0%, closely followed by Random Forest at 74.8%. The integration of XAI techniques provides actionable insights into key factors such as pipeline material, age, installation layout, and commodity type, which significantly influence shutdown durations. These findings underscore the practical implications of the proposed approach, enabling pipeline operators, emergency responders, and regulatory authorities to make informed decisions that optimize resource allocation and mitigate risks effectively.